Three Southeast Asian governments entered the AI race at almost the same time. Indonesia published its Strategi Nasional Kecerdasan Artifisial in 2020. Malaysia followed with its National Artificial Intelligence Roadmap 2021–2025. Vietnam issued Decision 127 on AI research, development, and application to 2030 in January 2021.
The documents read like variations on the same script: priority sectors, talent pipelines, ethics principles, economic ambition, and promises of national competitiveness. On paper, all looked serious enough. Five years later, they have moved along sharply different trajectories.
Vietnam has passed a standalone, risk-based AI law, approved in December 2025 and effective from March 2026. Malaysia has become one of Southeast Asia’s most aggressive magnets for data centre and cloud investment. Indonesia, the region’s largest economy, is still waiting for a presidential signature on an umbrella regulation guiding AI adoption across ministries and regional governments.
Their point of departure was broadly similar. The machinery was different. The real test of a national AI strategy comes after publication: in budget rooms, procurement decisions, institutional mandates, cloud contracts, and the slow discovery of who actually owns the work.
Stay ahead of the geopolitical week.
MD Briefing delivers expert analysis across five global fronts — the Indo-Pacific, energy, geoeconomics, European security, and the Middle East — every Monday morning. Free.
Indonesia: when the strategy has no spine
Indonesia moved early. Its 2020 AI strategy identified five priority sectors: health, bureaucratic reform, education and research, food security, and mobility and smart cities. It also recognised the importance of ethics, talent, data, infrastructure, and industrial innovation. What it lacked was a centre strong enough to turn those ideas into direction.
For years, Indonesia’s AI agenda moved without a binding umbrella regulation, a dedicated implementing agency, or a delivery structure capable of disciplining sectoral fragmentation. Reuters reported in 2025 that Indonesia was preparing a new AI roadmap, the first comprehensive AI document since a smaller ethics guideline in 2023. By June 2026, the draft presidential regulation was still awaiting President Prabowo Subianto’s approval.
In fast-moving technology policy, delay changes the field. Banks experiment. Start-ups adopt tools. Ministries launch pilots. Global technology firms enter the market. AI activity grows before AI governance matures.
The paradox is that Indonesia’s sectoral regulators have moved faster than the national framework meant to guide them. In April 2025, the Financial Services Authority launched Artificial Intelligence Governance for Indonesian Banks, giving the banking sector clearer supervisory expectations around responsible AI development than many other parts of the economy. The spokes are moving while the national umbrella still lags behind.
Indonesia’s lesson is blunt: a strategy can create conversation, but execution needs an empowered owner. A sovereign AI fund, fiscal incentives, and public-sector AI adoption may all help. Yet money only matters if it is attached to institutions that can spend it well.
Malaysia: when infrastructure arrives before capability
Malaysia shows a different pattern. It moved faster institutionally and commercially. In December 2024, it launched the National AI Office to centralise AI policy, strategic planning, research and development, and regulatory oversight.
Capital moved even faster. Government communications in 2025 referred to 143 approved data centre investment projects between 2021 and June 2025, worth RM144.4 billion in total. Of these, 25 had Malaysia Digital status under the Digital Ecosystem Acceleration Scheme, with reported expectations of 1,429 new jobs.
Can Malaysia turn hosting power into learning power?
Attracting AI infrastructure is an achievement. It shows that Malaysia has become strategically legible to global technology capital. But hosting data centres differs from building domestic AI capability. The harder task is producing engineers who design systems, firms that build applications and models, universities that train advanced talent at scale, and regulators that can steer infrastructure towards national learning.
Budget 2026 shows both ambition and constraint. Malaysia allocated RM18.1 million to the National AI Office and announced RM2 billion for a Sovereign AI Cloud. These are meaningful signals. Yet they sit beside private infrastructure commitments whose scale is far larger and whose logic is often shaped by hyperscaler strategy.
Malaysia has solved one problem many middle-income countries struggle with: attracting strategic infrastructure capital. The unresolved question is whether that capital will deepen Malaysian capability or make Malaysia an efficient host for someone else’s AI economy.
Grid constraints, water stress, and exposure to US export-control politics sharpen the issue. In the AI economy, land, energy, and connectivity matter. Sovereignty begins when infrastructure produces capability, bargaining power, and institutional learning.
Vietnam: when coherence creates its own risk
Vietnam’s trajectory is the most coherent of the three. Decision 127 set an early direction to 2030, and the country has since moved from strategy to law. Its AI Law, passed in December 2025 and effective from March 2026, establishes a standalone, risk-based framework for providers and deployers.
Vietnam is also moving towards more explicit targets. A draft updated national AI strategy released for public consultation in 2026 targets AI contributing around six per cent of GDP by 2030, the training of at least 500,000 workers in AI deployment, and domestic computing capacity equivalent to 250,000 advanced GPUs.
Vietnam’s distinction lies in the chain of command it has built around the strategy. Viettel has partnered with NVIDIA to develop a sovereign AI ecosystem and operates a cluster of 22 NVIDIA DGX B200 systems with reported performance of up to 1.5 exaFLOPs. FPT has collaborated with NVIDIA on Vietnamese persona datasets and AI factory infrastructure.
The result is alignment between legal direction, industrial champions, compute infrastructure, and national capability goals. Vietnam has treated AI as an instrument of state capacity, rather than only as a conventional technology sector.
That gives Vietnam speed, but also a different vulnerability. Capability now runs through a small number of state-linked conglomerates and one dominant foreign chip ecosystem. The model depends on export licensing, vendor roadmaps, supply chains, and the balance sheets of a few national champions. This may be a rational trade for speed. Vietnam has not escaped dependency; it has simply organised it better.
The missing actor in a multiplex digital ecosystem
Placed side by side, the three countries reveal three recurring failure modes in Global South AI strategies: the aspiration–execution gap, institutional orphanhood, and upstream dependency. Indonesia illustrates the first two most clearly, Malaysia the third, while Vietnam shows how stronger execution and clearer ownership can still create concentration risk.
The deeper issue is that most national AI strategies are still written as state-market bargains. The state sets direction. The market supplies capability. The document tries to negotiate between the two, but that map is no longer enough.
AI now operates inside what Amitav Acharya might call a multiplex world: many actors, no single director, and power distributed across institutions, firms, infrastructures, standards, and publics. A serious strategy for a multiplex digital ecosystem needs a third force: society with enough capacity to scrutinise both the state and the market.
Universities can test systems and train independent expertise. Technology journalists can surface failures before they become scandals. Civil society can defend data rights, bias scrutiny, and protection against automated exclusion. Citizens matter because the legitimacy of AI adoption depends on enforceable rights when systems make mistakes. This is a resilience function.
Each country’s weakness is precisely the kind of weakness an empowered third actor can help correct. In Indonesia, expert scrutiny can keep AI strategy alive across political cycles. In Malaysia, academic and civic scrutiny can ask whether infrastructure investment is producing real capability. In Vietnam, independent audit can act as a counterweight to concentrated champions.
This connects to a wider Global South problem. In my earlier essay, Claude Mythos and the Global South’s AI Governance Dilemma, I described the double asymmetry facing many developing countries: dependence on external technology providers, combined with limited influence over the rules, safety thresholds, and preparedness systems that govern those technologies. Southeast Asia shows how that asymmetry plays out domestically. A state may publish a strategy, attract capital, or designate champions, yet remain exposed if it cannot control the deeper conditions of capability.
What should change
The next generation of AI strategies in the Global South should begin less with aspiration and more with delivery.
First, strategies should be published together with implementing institutions. A roadmap without an empowered owner is vulnerable from the first day. Vietnam’s advantage lies in the alignment between law, institutions, and executors. Indonesia shows the cost of sequencing things the other way around.
Second, countries should measure dependency, not just infrastructure. Data centre capacity, project numbers, and foreign investment totals are useful, but they do not tell us who controls workloads, models, data, and high-value jobs. A more honest AI strategy would track domestic AI workloads, locally governed compute, national talent in high-value roles, and locally owned models in critical public functions.
Third, ASEAN should handle together what single countries cannot carry alone. Shared compute facilities, mutual recognition of model audits, common procurement standards, and regional AI safety exercises would give the region more leverage than fragmented national competition for data centre investment.
Fourth, AI strategies must be designed for a multiplex digital ecosystem. The state still matters. Markets still matter. Society should function as a supervisory actor with real capacity to intervene. Sovereign AI emerges when public institutions, domestic firms, universities, regulators, media, and citizens can collectively set and correct the terms under which AI operates.
The key lesson from Southeast Asia is that AI strategies rarely fail at launch, but later—when no institution owns delivery, infrastructure outpaces capability, and dependency is mistaken for sovereignty.
For the Global South, the next AI strategy should begin with a less glamorous question: who will have the authority, resources, and public accountability to make this real after the launch event is over?
